NuBoard

Matched top-layer HyperRAM routing candidate

Promoted to the main-board project on 2026-09-24. The authoritative PCB is now ../../nuBoard.kicad_pcb, with its schematic, pin-map CSV and displaced-signal FPGA constraints synchronized. This directory remains the verified pre-promotion reference snapshot; the description below records its original candidate status.

Verified 2026-09-24 with KiCad 10.0.6. Open nuBoard.kicad_pcb in this directory to inspect the result. This is a PCB-only candidate, separate from the main-board schematic and PCB. Do not update it from the unchanged main schematic: its pin allocation is different.

U2 and all 585 HyperRAM signal track segments (DQ0–7, CK, RWDS, CS# and RESET#) are locked against accidental editing. U2's power connections are still unrouted; shared board power networks were not locked.

Result

All ten required signals—CK, DQ0–7 and RWDS—are continuous, unbranched FPGA-pad-center to RAM-pad-center paths on F.Cu only, with zero vias. The PCB-only length spread is 0.214807 mm (8.457 mil), below the 0.381 mm (15 mil) limit by 0.166193 mm. No package compensation is configured. Unknown IC package skew is outside this approved requirement.

U2 remains on F.Cu at (90.8, 75.0) mm, 90°. No additional package movement or pin-map change from the last surface-routing trial was needed. CS# uses G3/A3 and RESET# uses G2/A4; their FPGA-to-RAM paths are also F.Cu-only without vias, but their pull-up branches remain unfinished.

The tuning patterns occupy the straight corridor between approximately x=93.3 and 99.0 mm. Coordinating their direction and spacing avoided the failed approach of inserting isolated large detours into crowded traces. CK and DQ0–7 target 11.0 mm; RWDS retains its original 11.214798 mm route. Trace width is 0.0762 mm, pattern period 0.44 mm, each bump's baseline width 0.22 mm, and corner bevel 0.01 mm. Bump counts for DQ0–7 are 9, 9, 2, 13, 3, 6, 10, 10 respectively; CK has one bump.

All 15 displaced signals retain the replacement routes documented in hyperram-top-layer-plan.md. Their nets have no unconnected items in the final DRC report.

Verification

• drc.json: fresh DRC with copper zones refilled and the copied main-board project/rules. No length-skew, shorting, clearance, hole-clearance or courtyard violations. Six dangling-track warnings remain outside the fast group, along with 27 whole-board unconnected items.
• verification.json: independent pad-to-pad graph/length audit, with SHA-256 hashes of the checked PCB and DRC report. It checks both endpoint identities, DDIO capability, RWDS clock capability, absence of branches, disconnected copper and self-crossings, F.Cu-only routing, zero vias, unset package compensation, and the ten-net maximum length spread.
• validation/test_hyperram_surface.py: mutation tests reject an open segment, a via, a bottom-layer segment, excessive length, a dangling branch and package compensation, even with the previously clean DRC.

From the repository root, repeat:

kicad-cli pcb drc --refill-zones --format json \
  -o hardware/main-board/routing/hyperram-top/drc.json \
  hardware/main-board/routing/hyperram-top/nuBoard.kicad_pcb

# Use KiCad's Python environment (pcbnew and shapely are required).
python3 hardware/main-board/validation/check_hyperram_surface.py \
  hardware/main-board/routing/hyperram-top/nuBoard.kicad_pcb \
  --drc hardware/main-board/routing/hyperram-top/drc.json
python3 hardware/main-board/validation/test_hyperram_surface.py

On this Mac the Python executable used was /Applications/KiCad/KiCad.app/Contents/Frameworks/Python.framework/Versions/3.9/bin/python3.

Not full-board signoff

The requested ten-signal routing/length objective is met in this candidate. The whole board is not fabrication-ready. Power/bypass routing, CS#/RESET# pull-up connections, existing unrelated open connections, and main-project schematic/CSV/FPGA-constraint synchronization remain to do. The main project has deliberately not been overwritten by this candidate.

Geometric length matching is not a signal-integrity simulation or Efinity timing closure. The allocation spans clock regions L3–L6; validate RWDS capture phase/distribution, DDIO implementation, and timing with the actual HyperRAM controller. The existing APS1604M controller remains incompatible.